Photovoltaic screw pile position deviation detection method, system and terminal

Abstract

The invention relates to the technical field of engineering intelligent detection, and particularly discloses a photovoltaic screw pile position deviation detection method, system and terminal, and the method comprises the steps: carrying out the target recognition and feature extraction of visual image data through a target detection module based on deep learning, and carrying out the filtering and feature extraction of laser radar data, meanwhile, a multi-source data fusion technology is combined to cooperatively utilize visual and laser radar data to provide richer positioning information for coordinate recognition of the screw pile, finally, pile position deviation automatic detection is achieved based on the positioning information, and concrete and accurate deviation detection results are provided for a construction team; therefore, effective data support is provided for subsequent rectification and adjustment, and the reworking cost and time are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engineering intelligent detection, and more specifically, to a photovoltaic screw pile position deviation detection method, system and terminal. BACKGROUND

[0002] In the construction process of a photovoltaic project, as the foundation of a photovoltaic support, the accuracy of the installation position of a screw pile directly affects the stability and power generation efficiency of the photovoltaic system.

[0003] Traditional screw pile position detection mainly relies on manual measurement or fixed equipment detection, which has obvious shortcomings in large-area photovoltaic projects. For example, in a large photovoltaic power station, manual pile-by-pile detection is time-consuming and labor-intensive, and cannot meet the rapid construction needs of large-scale photovoltaic projects. More specifically, a 100MW photovoltaic power station may contain tens of thousands of screw piles, and manual detection may take several weeks or even months.

[0004] For another example, some detection systems based on fixed laser scanning equipment have high precision, but have high installation and maintenance costs, and are difficult to adapt to the needs of complex terrain and large-area photovoltaic projects.

[0005] Therefore, there is an urgent need for an efficient and automated photovoltaic screw pile position deviation detection scheme. SUMMARY

[0006] The technical problem to be solved by the present application is to provide a photovoltaic screw pile position deviation detection method, system and terminal, which solves the problem of low efficiency and high cost caused by relying on manual measurement or fixed equipment for screw pile position deviation detection in the prior art.

[0007] The technical problem to be solved by the present application is solved by the following technical solution: In a first aspect, the present application provides a photovoltaic screw pile position deviation detection method, comprising: obtaining original laser radar data and original visual image data, the original laser radar data and the original visual image data being collected by a UAV during flight along a preset path, the UAV being provided with a camera; performing data quality monitoring on the original laser radar data and the original visual image data, and triggering a re-sampling mechanism if there is data that does not meet the preset requirements until all data meets the preset requirements, to obtain laser radar data and visual image data; using a target detection module based on deep learning to perform target recognition and feature extraction on the visual image data to obtain first center point coordinates; performing filtering and feature extraction on the laser radar data to obtain second center point coordinates; performing multi-source data fusion on the first center point coordinates and the second center point coordinates to obtain first screw pile center point coordinates; The pile position deviation detection is performed based on the first spiral pile center point coordinates to obtain a deviation detection result.

[0008] Further, data quality monitoring is performed on the original laser radar data and the original visual image data, and if there is data that does not meet the preset requirements, a resampling mechanism is triggered until all data meet the preset requirements, thereby obtaining the laser radar data and the visual image data, including: An image blurriness of the original visual image data is calculated, and when the image blurriness is greater than a preset blurring threshold, the exposure time of the camera is adjusted and a resampling mechanism is triggered to obtain new original visual image data, wherein the image blurriness is obtained by calculating a Laplacian operator.

[0009] Further, a target detection module based on deep learning is used to perform target recognition and feature extraction on the visual image data to obtain first center point coordinates, including: Image correction and image enhancement are performed on the visual image data to obtain enhanced visual image data; A spiral pile recognition network based on a YOLOv7 algorithm is used to perform target detection on the enhanced visual image data to obtain a spiral pile recognition image, wherein the backbone network of the spiral pile recognition network based on the YOLOv7 algorithm embeds a channel attention mechanism, and a CIoU Loss loss function is used in the training process; The contour in the spiral pile recognition image is extracted, and the first center point coordinates are determined by calculating the minimum circumscribed circle.

[0010] Further, filtering and feature extraction are performed on the laser radar data to obtain second center point coordinates, including: The laser radar data is cropped by a region of interest to obtain cropped laser radar data; A voxel grid-based filtering method is used to process the cropped laser radar data to obtain denoised laser radar data; A random sample consensus algorithm is used to fit a cylindrical surface to the denoised laser radar data, wherein the second center point coordinates can be determined in the cylindrical surface fitting process.

[0011] Further, multi-source data fusion is performed on the first center point coordinates and the second center point coordinates to obtain first spiral pile center point coordinates, including: The first center point coordinates and the second center point coordinates are converted into first center point coordinate vectors and second center point coordinate vectors; A first covariance matrix representing visual positioning reliability and a second covariance matrix representing laser radar positioning reliability are obtained; The first center point coordinate vectors and the second center point coordinate vectors are fused based on the first covariance matrix and the second covariance matrix to obtain first spiral pile center point coordinate vectors according to the following formula:

[0012] wherein, is a first spiral pile center point coordinate vector, P1 and P2 are respectively a first covariance matrix and a second covariance matrix, and are respectively a first center point coordinate vector and a second center point coordinate vector; extracting a three-dimensional coordinate value from the first spiral pile center point coordinate vector; obtaining a design center point coordinate of the first spiral pile; calculating a three-dimensional deviation evaluation value of the three-dimensional coordinate value based on the design center point coordinate, and taking the three-dimensional coordinate value as the first spiral pile center point coordinate when the three-dimensional deviation evaluation value is less than a preset deviation threshold.

[0013] Further, based on the first spiral pile center point coordinate, a pile position deviation detection is performed to obtain a deviation detection result, including: calculating a plane deviation value and an elevation deviation value based on the design center point coordinate and the first spiral pile center point coordinate according to the following formula;

[0014] wherein, E XY is a plane deviation value, (X real ,Y real ,Z real ) is the first spiral pile center point coordinate, (X des ,Y des ,Z des ) is the design center point coordinate, and E Z is an elevation deviation value; outputting the three-dimensional deviation evaluation value, the plane deviation value and the elevation deviation value as the deviation detection result.

[0015] Further, based on the first spiral pile center point coordinate, a pile position deviation detection is performed to obtain a deviation detection result, including: obtaining a second spiral pile center point coordinate of a second spiral pile adjacent to the first spiral pile; obtaining design center point coordinates of the first spiral pile and the second spiral pile respectively; calculating a spacing deviation value of the first spiral pile and the second spiral pile according to the following formula;

[0016] wherein, D des,ij is a design spacing of the first spiral pile and the second spiral pile, D real,ij is an actual spacing of the first spiral pile and the second spiral pile, is a spacing deviation value, X des,i and YX and Y are respectively the x-axis and y-axis coordinate values of the design center point coordinates of the first screw pile, des,j and X des,j and Y real,i are respectively the x-axis and y-axis coordinate values of the design center point coordinates of the second screw pile, real,i X and Y are respectively the x-axis and y-axis coordinate values of the first screw pile center point coordinates, real,j and X real,j and Y G are respectively the x-axis and y-axis coordinate values of the second screw pile center point coordinates; The spacing deviation value is output as a deviation detection result.

[0017] In a second aspect, the present application provides a photovoltaic screw pile position deviation detection system, which is used to execute the method provided by the first aspect of the present application or any possible implementation manner of the first aspect of the present application.

[0018] In a third aspect, the present application provides a terminal, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the method provided by the first aspect of the present application or any possible implementation manner of the first aspect of the present application.

[0019] In summary, the present application includes at least one of the following beneficial technical effects: By combining deep learning and multi-source data fusion technology, more rich positioning information is provided by collaborative use of visual and laser radar data, the limitations of single sensor data are overcome, data quality monitoring processing is introduced, and the collected data is ensured to meet the preset requirements by combining the resampling mechanism, so as to ensure the reliability of subsequent calculation and analysis, reduce the false detection or missed detection caused by data quality problems, and improve the stability of the detection scheme; The unmanned aerial vehicle equipped with multiple sensors is used for automatic data collection, which can quickly cover a large area to be detected, and avoid the problems of low efficiency and difficult terrain adaptation of traditional manual detection or fixed equipment detection; The recognized coordinates are converted into multi-dimensional, intuitive and action-guiding engineering quality data, realizing efficient detection of photovoltaic screw pile position deviation, providing concrete and accurate deviation detection results for the construction team, providing effective data support for subsequent rectification and adjustment, and reducing rework cost and time; The introduction of spacing deviation value calculation expands from single deviation detection to group relationship detection, which directly reflects the deviation degree between adjacent screw piles, and complements the single deviation detection, together forming a powerful deviation detection system. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 FIG. 1 is a flowchart of a photovoltaic screw pile position deviation detection method; Figure 2 A visual schematic diagram for target detection and deviation detection in embodiments of the present application. DETAILED DESCRIPTION

[0021] In order to facilitate understanding of the technical means, creative features, purposes and effects of the present application, the present application will be further described below in combination with specific drawings.

[0022] The present application discloses a photovoltaic spiral pile position deviation detection method, which first acquires original laser radar data and original visual image data, and performs data quality monitoring on the original laser radar data and the original visual image data, so as to ensure that the collected data meets the use requirements of the subsequent model and ensures the reliability of analysis and calculation. Then, a target detection module based on deep learning is used to perform target recognition and feature extraction on the visual image data to obtain first center point coordinates, and laser radar data is filtered and feature extracted to obtain second center point coordinates, so as to provide more rich positioning information for coordinate recognition of the spiral pile by cooperatively using visual and laser radar data. Then, the first center point coordinates and the second center point coordinates are fused to obtain first spiral pile center point coordinates; and based on the first spiral pile center point coordinates, pile position deviation detection is performed to obtain a deviation detection result.

[0023] In this way, by combining deep learning and multi-source data fusion technology, the rich positioning information in visual and laser radar data is cooperatively used to identify the center of the spiral pile and further calculate the coordinate information thereof, the recognized coordinates are converted into multi-dimensional, intuitive and action-guiding engineering quality data, efficient detection of the photovoltaic spiral pile position deviation is realized, specific and accurate deviation detection results are provided for the construction team, effective data support is provided for subsequent rectification and adjustment, and the cost and time of rework are reduced; at the same time, the group relationship detection and the single deviation detection are complementary to each other, and a powerful deviation detection system is jointly constructed, and the automatic detection scheme for the photovoltaic spiral pile position deviation is improved.

[0024] As shown in Figure 1 The photovoltaic spiral pile position deviation detection method provided by the present application specifically includes the following steps: S1, acquiring original laser radar data and original visual image data, the original laser radar data and the original visual image data being collected by a UAV during flight on a preset path, the UAV being provided with a camera.

[0025] In one example, the UAV with vertical take-off and landing design can operate flexibly in narrow or complex terrain, while having the characteristics of long endurance and high stability. Among them, the UAV carries multiple sensors, including but not limited to Lidar module for obtaining high-precision raw Lidar data (three-dimensional point cloud data), camera module for collecting raw visual image data (visual image of screw pile), such as high-definition camera, etc. In addition, the GPS / IMU (Global Positioning System / Inertial Measurement Unit) module is used to provide the position, attitude and speed information of the UAV to realize accurate flight control. Other auxiliary sensors such as barometer, compass, etc. are used to enhance flight stability and environmental perception ability.

[0026] Before detection, according to the layout and terrain characteristics of the photovoltaic project, combined with the design drawings, the flight path is planned above each screw pile construction point. The goal of flight path planning is to ensure that the UAV can cover all the screw piles to be detected, while avoiding obstacles, improving flight efficiency and data acquisition quality.

[0027] The UAV flies according to the preset path, while activating various sensors to collect data. The data collection process includes: 1. Raw Lidar data collection: "Distance trigger" mode is used, when the distance between the UAV and the screw pile design coordinate is ≤10m, the Lidar is started (scanning frequency 20Hz), and 2 seconds (about 40 frames of point cloud) are continuously collected. The amount of single-frame point cloud data is about 10MB. The point cloud collection range focuses on 0-1m above the ground (covering the screw pile unearthed height of 20-50cm).

[0028] 2. Raw visual image data collection: "Position trigger" mode is used, when the UAV reaches the design coordinate of the screw pile directly above (error ≤2m), the camera module is triggered to take pictures, and each image is attached with GPS timestamp and attitude information.

[0029] 3. Flight state data collection: GPS / IMU data is continuously collected (sampling rate 200Hz), each frame of data contains position: X G ,Y G ,Z G (RTK positioning result); speed: v x ,v y ,v z (IMU solution); attitude: φ (roll angle), θ (pitch angle); acceleration: a x ,a y ,a z (IMU raw data), etc. Among them, the flight state data can provide the real-time flight state of the UAV, and provide important positioning information for the collection mode of raw Lidar data and raw visual image data, to ensure the smooth collection of important data.

[0030] In step S1, the unmanned aerial vehicle flies along the preset path, and laser radar point cloud data and visual image data are collected synchronously, which overcomes the low efficiency and terrain adaptability problems of traditional manual detection or fixed equipment detection.

[0031] S2, data quality monitoring is performed on the original laser radar data and the original visual image data, and if there is data that does not meet the preset requirements, a resampling mechanism is triggered until all data meet the preset requirements, and laser radar data and visual image data are obtained.

[0032] In one example, the point cloud density of the original laser radar data is detected, and when the point cloud density is <100 points / m 2 , the flight speed is reduced (to 2 m / s) and the resampling mechanism is triggered to collect new original laser radar data.

[0033] In one example, the image blur of the original visual image data is calculated, and when the image blur is greater than a preset blur threshold, the exposure time of the camera is adjusted and the resampling mechanism is triggered to obtain new original visual image data, wherein the image blur is calculated by calculating the Laplacian operator.

[0034] More specifically, the image blur Blur is calculated using the following formula, and when the image blur Blur is greater than 0.1, the exposure time of the camera is adjusted and the resampling mechanism is triggered to obtain new original visual image data;

[0035] where W and H represent the width and height of the original visual image data, respectively, I(x,y) represents the pixel value at coordinates (x,y) in the original visual image data, represents the calculation of the Laplacian operator.

[0036] Here, the Laplacian operator is used to quantify the blur of the original visual image data, and then the visual image data with high blur is removed to ensure that the visual image data input to the subsequent model has a certain degree of clarity and improves the recognition accuracy of the model.

[0037] In step S2, the data quality monitoring process is introduced, and the resampling mechanism is combined to ensure that the collected data meets the preset requirements, thereby ensuring the reliability of subsequent calculation and analysis, reducing false detection or missed detection caused by data quality problems, and improving the stability of the detection scheme.

[0038] S3, using a target detection module based on deep learning to identify and extract features of the visual image data to obtain the first center point coordinates.

[0039] Specifically, the visual image data is first enhanced through image correction and image enhancement. Image correction includes geometric correction and grayscale correction. Geometric correction, such as distortion correction and projection transformation, can eliminate geometric distortions in the visual image data caused by camera lens distortion and shooting angle. In one example, Zhang's calibration method is used for camera calibration to establish a camera model and achieve accurate geometric correction. Grayscale correction adjusts the brightness, contrast, and color balance of the visual image data, improving the visual effect and feature recognition of the image. In one example, a combination of adaptive histogram equalization and gamma correction is used to enhance the contrast and detail of the image.

[0040] Image enhancement includes noise filtering and edge enhancement. In one example, methods such as bilateral filtering or median filtering are used to remove noise from the image while preserving its edges and details; and edge enhancement, such as using the Sobel operator, highlights the outline of the helical studs, facilitating subsequent target detection and localization.

[0041] Next, a deep learning-based object detection module is used to perform object recognition on the visual image data. In a specific example, a helical pile recognition network based on the YOLOv7 algorithm is used to perform object detection on the enhanced visual image data to obtain a helical pile recognition image.

[0042] YOLOv7 treats object detection as a regression problem, directly predicting bounding boxes and class probabilities on the image (e.g., ...). Figure 2 (As shown). Specifically, the helical post recognition network based on the YOLOv7 algorithm embeds a channel attention mechanism in its backbone network, which enhances its feature extraction capability for the helical post region. The channel attention mechanism can be expressed by the formula: CA(F)=Conv(Softmax(FC(GlobalAvgPool(F)))) F in, represents element-wise multiplication, F represents the feature map, GlobalAvgPool represents global mean pooling, FC represents a fully connected layer, Softmax represents the Softmax activation function, Conv represents convolution, and CA represents CoordAttention channel attention enhancement.

[0043] In the channel attention mechanism, feature information highly correlated with the helical pile features is extracted from the original feature map and transformed into weights applied to the target feature, thereby enhancing the target feature. Furthermore, a four-branch pooling method (1×1, 3×3, 5×5, 7×7) is employed in the spatial pyramid pooling part to improve adaptability to helical piles of different scales. That is: SPP(F) = Concat(MaxPool(F, 1), MaxPool(F, 3), MaxPool(F, 5), MaxPool(F, 7)) Wherein, SPP(F) represents the spatial pyramid pooling processing of the feature map, Concat represents concatenation, MaxPool represents maximum pooling, 1, 3, 5, 7 all represent the pooling size.

[0044] In the model training process, CIoU Loss loss function is used for optimization to make the boundary box regression more accurate. The CIoU Loss loss function is expressed by the formula:

[0045] Wherein, CIoU represents the CIoU Loss loss function value, IoU represents the intersection over union, that is, the IoU(Intersection over Union) value, p is the distance between the predicted box b and the real box b gt , c is the length of the diagonal line of the minimum rectangle surrounding the predicted box and the real box, a is the balance coefficient, and v is the aspect ratio consistency parameter.

[0046] The CIoU Loss loss function not only considers the overlapping area during training, but also considers the center point distance and the difference in aspect ratio, so that the model is directly guided to optimize the positioning accuracy of the boundary box during training, and outputs a boundary box that is more consistent with the real position during prediction.

[0047] Finally, the contour in the spiral pile recognition image is extracted, and the first center point coordinates are determined by calculating the minimum circumscribed circle. Here, the boundary box output by the spiral pile recognition network based on the YOLOv7 algorithm already contains a rough center point, but direct use may have insufficient accuracy. Since the top of the spiral pile is circular, compared with the rectangular boundary box, the minimum circumscribed circle is more consistent with the true geometry of the target. By extracting the pixel-level contour and calculating the minimum circumscribed circle to determine the first center point coordinates (X v , Y v ), a higher visual measurement accuracy can be achieved.

[0048] S4, filtering and feature extraction are performed on the laser radar data to obtain second center point coordinates.

[0049] Specifically, first, the laser radar data is cropped by a region of interest (ROI) to reduce the amount of invalid data (cropping the part not containing the screw pile, such as point cloud data with a height greater than 1 meter), to obtain cropped laser radar data; then, a voxel grid-based filtering method is used to process the cropped laser radar data to obtain denoised laser radar data; then, a random sample consensus (RANSAC) algorithm is used to fit a cylindrical surface to the denoised laser radar data.

[0050] where the cylindrical surface model formula is: (X-X L ) 2 +(Y-Y L ) 2 =(r L ) 2 ,Z Lmin ≤Z≤Z Lmax . Here, (X L ,Y L ) is the projection of the cylindrical center in the XY plane, r L is the radius, and Z Lmin to Z Lmax is the height range of the cylinder (screw pile). In the above process, the error is also calculated using the following formula :

[0051] where M is the number of inliers, (X p ,Y p ) is the coordinate of the inlier p, (X L ,Y L ) is the center of the cylinder, and r L is the radius. When the error value is <0.001 m 2 , it is considered a valid fit.

[0052] In the above cylindrical surface fitting process, the second center point coordinates (X L ,Y L ,Z L ) can be determined, where Z L= (Z Lmin +Z Lmax ) / 2.

[0053] In step S4, the region of interest is cropped to remove a large number of background points irrelevant to the screw pile, reducing the burden of subsequent algorithms; the filtering method based on voxel grid can remove noise points caused by sensor errors, air suspensions and other reasons, and the center or barycenter of the divided region is used to represent the voxel, so as to effectively remove noise while better preserving the macroscopic geometric structure of the original point cloud; the random sample consensus algorithm uses the physical model of the screw pile, i.e. the cylindrical surface model, to calculate the center point coordinates from the three-dimensional point cloud, realizing accurate positioning of the screw pile.

[0054] S5, multi-source data fusion is performed on the first center point coordinates and the second center point coordinates to obtain first screw pile center point coordinates.

[0055] Specifically, the first center point coordinates and the second center point coordinates are first converted into first center point coordinate vectors and second center point coordinate vectors. It is worth mentioning that the coordinate conversion step is also included before conversion into vector form, so that the first center point coordinate vector and the second center point coordinate vector have a common coordinate system.

[0056] In one example, the pixel coordinates (X v ,Y v ) obtained based on image recognition are converted into geodetic coordinates (X v ’,Y v ’); at the same time, the second center point coordinates (X L ,Y L ,Z L ) are converted into the same coordinate system, respectively obtaining the first center point coordinate vector =(X v ’,Y v ’,Z v ’) and the second center point coordinate vector =(X L ’,Y L ’,Z L ’), wherein Z v ’=Z L ’.

[0057] At the same time, the first covariance matrix representing the visual positioning reliability and the second covariance matrix representing the laser radar positioning reliability are obtained.

[0058] In one example, the preset first covariance matrix and the second covariance matrix are obtained through offline calibration. That is, the first covariance matrix and the second covariance matrix are determined based on the prior calibration results of the camera and the lidar. By analyzing the repeated measurement errors of the sensors on the known position calibration objects, the covariance matrix of the positioning results is calculated as a fixed prior value. Specifically, by shooting calibration boards with known accurate coordinates at different positions, the distribution of the visual positioning errors is counted to form the first covariance matrix; and by scanning target balls with known accurate coordinates, the distribution of the lidar positioning errors is counted to form the second covariance matrix.

[0059] In one example, the first covariance matrix and the second covariance matrix are obtained based on the principle of federal Kalman filter (FKF). That is, the first covariance matrix and the second covariance matrix are updated synchronously at each time of updating the state, and are called in the next fusion process. Wherein, the initial state vector x0 of the federal Kalman filter is defined as: and the initial covariance matrix P0 is defined as:

[0060] Wherein, P1 and P2 are the first covariance matrix and the second covariance matrix respectively.

[0061] The observation value of the first sub-filter is the geodetic coordinate converted from the pixel coordinate of the image-recognized screw pile, and the observation equation is:

[0062] The observation value of the second sub-filter is the cylindrical surface parameter fitted from the point cloud, and the observation equation is:

[0063] Wherein, v Vision is the observation noise of the first sub-filter, X, Y and Z are the three-dimensional coordinate values of the current observation, Z Vision is the observation quantity of the first sub-filter, θ is the circumferential angle of the cylindrical section, t is the height direction parameter, v LiDAR is the observation noise of the second sub-filter, and Z LiDAR is the observation quantity of the second sub-filter.

[0064] Then, based on the first covariance matrix and the second covariance matrix, the first screw pile center point coordinate vector is obtained by fusing the first center point coordinate vector and the second center point coordinate vector according to the following formula:

[0065] Wherein, is the first screw pile center point coordinate vector, P1 and P2 are the first covariance matrix and the second covariance matrix respectively, and respectively, are the first and second center point coordinate vectors.

[0066] Then, three-dimensional coordinate values (X real ,Y real ,Z real ) are extracted from the first spiral pile center point coordinate vector; and design center point coordinates (X des ,Y des ,Z des ) of the first spiral pile are obtained. The design center point coordinates are directly extracted from design drawings or CAD models, or manually input by a user.

[0067] Finally, a three-dimensional deviation evaluation value of the three-dimensional coordinate values is calculated based on the design center point coordinates in the following formula:

[0068] wherein E is the three-dimensional deviation evaluation value, (X real ,Y real ,Z real ) are the three-dimensional coordinate values extracted from the first spiral pile center point coordinate vector, and (X des ,Y des ,Z des ) are the design center point coordinates.

[0069] In the technical solution of the present application, when the three-dimensional deviation evaluation value is less than a preset deviation threshold, the three-dimensional coordinate values are taken as the first spiral pile center point coordinates.

[0070] In the above step S5, the first and second covariance matrices respectively describe the reliabilities of the visual positioning and the laser radar positioning, and the first and second center point coordinates are fused based on the reliabilities as weights, so that the uncertainties of the visual data and the laser radar data for the center point positioning can be dynamically processed, so that more reliable data is given a greater proportion in the fusion process, and a most optimal and least uncertain fusion result is further generated.

[0071] In addition, the three-dimensional deviation evaluation value is set as a quality threshold to ensure that the first spiral pile center point coordinates do not deviate from the actual values, and to further prevent the positioning data with obvious errors from being taken as the input of the deviation detection, thereby reducing the reliability.

[0072] S6, performing pile position deviation detection based on the first spiral pile center point coordinates to obtain a deviation detection result.

[0073] In one example, planar deviation values and elevation deviation values are calculated based on the design center point coordinates and the first spiral pile center point coordinates in the following formula:

[0074] wherein E XY is a planar deviation value, (X real , Y real , Z real ) is a first helical pile center point coordinate, (X des , Y des , Z des ) is a design center point coordinate, E Z is an elevation deviation value; The three-dimensional deviation evaluation value, the planar deviation value and the elevation deviation value are output as deviation detection results.

[0075] It can be understood that using the planar deviation value and the elevation deviation value to calculate converts the identified first helical pile center point coordinate into multi-dimensional, intuitive and action-guiding engineering quality data, which can provide a concrete and accurate deviation detection result for the construction team.

[0076] Wherein, using the Euclidean distance formula to calculate the planar deviation value can reflect the deviation distance of the center of the helical pile in the horizontal plane from the design position. And using the absolute value to calculate the difference in height reflects whether the top of the helical pile is on the same horizontal plane, which is related to the flatness and structural stress uniformity after the installation of the photovoltaic panel. This multi-dimensional deviation result output makes the subsequent rectification measures more accurate and efficient, greatly reducing the rework cost and time, and realizing the automation, precision and intelligence of the helical pile position detection.

[0077] In one example, first, the second helical pile center point coordinate of a second helical pile adjacent to the first helical pile is obtained; at the same time, the design center point coordinates of the first helical pile and the second helical pile are obtained respectively; then, the spacing deviation value of the first helical pile and the second helical pile is calculated by the following formula:

[0078] wherein D des,ij is a design spacing of the first helical pile and the second helical pile, D real,ij is an actual spacing of the first helical pile and the second helical pile, is a spacing deviation value, X des,i and Y des,i are x-axis and y-axis coordinate values of the design center point coordinate of the first helical pile, X des,j and Y des,j are x-axis and y-axis coordinate values of the design center point coordinate of the second helical pile, X real,i and Y real,i are x-axis and y-axis coordinate values of the first helical pile center point coordinate, X real,j and Y real,j are x-axis and y-axis coordinate values of the second helical pile center point coordinate; Finally, the interval deviation value is output as a deviation detection result.

[0079] In the above technical solution, the interval deviation value calculation is introduced to extend from single deviation detection to group relationship detection, directly reflecting the deviation degree between adjacent spiral piles. This measurement method can identify the abnormal state that the photovoltaic support cannot be installed due to the excessive pile position deviation of the spiral pile and the spiral pile in the case of small individual deviation degree.

[0080] At the same time, the interval deviation itself as a relative measurement, its focus is on the relative relationship between two spiral piles, so it can be immune to part of the local system error in the global coordinate system. This measurement method is complementary to single deviation detection, and together they form a more powerful deviation detection system.

[0081] The present application also discloses a photovoltaic spiral pile pile position deviation detection system for executing the above photovoltaic spiral pile pile position deviation detection method.

[0082] The present application also discloses a terminal including a processor, a memory, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the steps in the above various photovoltaic spiral pile pile position deviation detection method embodiments.

[0083] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. One or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the terminal device.

[0084] The terminal device can be a desktop computer, a notebook, a palm computer, and a cloud server, etc. The terminal device can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the examples of the terminal device do not constitute a limitation, and can include more or fewer components, or combine certain components, or different components, for example, the terminal device can also include an input / output device, a network access device, a bus, etc.

[0085] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the terminal device, and connects all parts of the terminal device through various interfaces and lines.

[0086] The memory can be used to store computer programs and / or modules, and the processor realizes various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to use of the terminal device (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.

[0087] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, and all these changes and improvements fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A photovoltaic helical pile deviation detection method, characterized in that, The method comprises the following steps: acquiring original laser radar data and original visual image data collected by a UAV during flight along a preset path, the UAV being provided with a camera; monitoring the data quality of the original laser radar data and the original visual image data, triggering a re-sampling mechanism if there is data that does not meet the preset requirements until all data meets the preset requirements, to obtain laser radar data and visual image data; using a deep learning-based target detection module to perform target recognition and feature extraction on the visual image data to obtain first center point coordinates; filtering and feature extraction on the laser radar data to obtain second center point coordinates; multi-source data fusion on the first center point coordinates and the second center point coordinates to obtain first screw pile center point coordinates; screw pile position deviation detection based on the first screw pile center point coordinates to obtain a deviation detection result.

2. The photovoltaic helical pile position deviation detection method according to claim 1, characterized in that, The method for monitoring the data quality of the original laser radar data and the original visual image data, triggering a re-sampling mechanism if there is data that does not meet the preset requirements until all data meets the preset requirements to obtain laser radar data and visual image data, comprises: calculating the image blurriness of the original visual image data, adjusting the exposure time of the camera and triggering a re-sampling mechanism to obtain new original visual image data when the image blurriness is greater than a preset blurring threshold, wherein the image blurriness is obtained by calculating the Laplacian operator.

3. The photovoltaic helical pile position deviation detection method according to claim 1, characterized by, The method for using a deep learning-based target detection module to perform target recognition and feature extraction on the visual image data to obtain first center point coordinates comprises: image correction and image enhancement on the visual image data to obtain enhanced visual image data; using a YOLOv7 algorithm-based screw pile recognition network to perform target detection on the enhanced visual image data to obtain a screw pile recognition image, wherein the backbone network of the YOLOv7 algorithm-based screw pile recognition network is embedded with a channel attention mechanism, and a CIoU Loss loss function is used during training; extracting the contour in the screw pile recognition image to determine the first center point coordinates by calculating the minimum circumscribed circle.

4. The photovoltaic helical pile position deviation detection method according to claim 1, characterized by, The method for filtering and feature extraction on the laser radar data to obtain second center point coordinates comprises: cropping the laser radar data by a region of interest to obtain cropped laser radar data; using a voxel grid-based filtering method to process the cropped laser radar data to obtain denoised laser radar data; using a random sample consensus algorithm to perform cylindrical surface fitting on the denoised laser radar data, wherein the second center point coordinates can be determined during cylindrical surface fitting.

5. The photovoltaic helical pile position deviation detection method according to claim 1, characterized by, The method for multi-source data fusion on the first center point coordinates and the second center point coordinates to obtain first screw pile center point coordinates comprises: converting the first center point coordinates and the second center point coordinates into first center point coordinate vectors and second center point coordinate vectors; obtaining a first covariance matrix representing visual positioning reliability and a second covariance matrix representing laser radar positioning reliability; fusing the first center point coordinate vector and the second center point coordinate vector based on the first covariance matrix and the second covariance matrix to obtain a first spiral pile center point coordinate vector according to the following formula: wherein, is the first center point coordinate vector of the first pile, and P1 and P2 are the first covariance matrix and the second covariance matrix, respectively, and are the first center point coordinate vector and the second center point coordinate vector, respectively. extracting a three-dimensional coordinate value from the first spiral pile center point coordinate vector; obtaining a design center point coordinate of a first spiral pile; calculating a three-dimensional deviation evaluation value of the three-dimensional coordinate value based on the design center point coordinate, and taking the three-dimensional coordinate value as the first spiral pile center point coordinate when the three-dimensional deviation evaluation value is less than a preset deviation threshold.

6. The photovoltaic helical pile position deviation detection method according to claim 5, characterized by, performing pile position deviation detection based on the first spiral pile center point coordinate to obtain a deviation detection result, including: calculating a planar deviation value and an elevation deviation value based on the design center point coordinate and the first spiral pile center point coordinate according to the following formula: wherein E XY is the planimetric deviation value, (X real , Y real , Z real ) is the first spiral pile center point coordinate, (X des , Y des , Z des ) is the design center point coordinate, E Z is the elevation deviation value; outputting the three-dimensional deviation evaluation value, the planar deviation value and the elevation deviation value as the deviation detection result.

7. The photovoltaic helical pile position deviation detection method according to claim 1, characterized by, performing pile position deviation detection based on the first spiral pile center point coordinate to obtain a deviation detection result, including: obtaining a second spiral pile center point coordinate of a second spiral pile adjacent to the first spiral pile; obtaining design center point coordinates of the first spiral pile and the second spiral pile respectively; calculating a spacing deviation value of the first spiral pile and the second spiral pile according to the following formula; wherein D des,ij is the designed distance between the first screw pile and the second screw pile, D real,ij is the actual distance between the first screw pile and the second screw pile, is the distance deviation value, X des,i and Y des,i are the x-axis and y-axis coordinate values of the designed center point coordinate of the first screw pile, X des,j and Y des,j are the x-axis and y-axis coordinate values of the designed center point coordinate of the second screw pile, X real,i and Y real,i are the x-axis and y-axis coordinate values of the center point coordinate of the first screw pile, X real,j and Y real,j are the x-axis and y-axis coordinate values of the center point coordinate of the second screw pile. outputting the spacing deviation value as the deviation detection result.

8. A photovoltaic monopile pile position deviation detection system, characterized in that, A computer program product for performing the method of any one of claims 1-7.

9. A terminal, characterized by comprising: A computer program product for performing the method of any one of claims 1-7, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method of any one of claims 1-7 when executing the computer program.

Images